Establishing the macular grading grid by means of fovea centre detection using anatomical-based and visual-based...

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate macular grading in digital retinal images is essential for diagnosing and monitoring eye diseases.
• Establishing a reliable macular grading grid relies heavily on precise fovea centre detection.
• Current methods may face challenges with image quality and anatomical variations.

Purpose of the Study:

• To develop and validate a robust methodology for fovea centre detection in digital retinal images.
• To establish an accurate macular grading grid using the detected fovea centre.
• To combine visual and anatomical features for improved fovea localisation.

Main Methods:

• A novel approach combining a priori anatomical features (optic disc, vascular tree) with visual criteria for fovea centre estimation.
• Utilisation of morphological processing to refine fovea centre detection when the fovea is visible.
• Handling cases where the fovea is indistinguishable by retaining the initial estimation.
• Validation on the MESSIDOR and DIARETDB1 public retinal image databases.

Main Results:

• The methodology achieved high accuracy in fovea centre detection across two independent datasets.
• 98.24% of fovea centres were accurately detected within the Excellent to Fair categories in the MESSIDOR database.
• 94.38% of fovea centres were accurately detected within the Excellent to Fair categories in the DIARETDB1 database.
• The combined feature approach demonstrated superior performance in fovea localisation.

Conclusions:

• The proposed methodology provides a reliable and accurate method for fovea centre detection in digital retinal images.
• This technique facilitates the establishment of a precise macular grading grid, aiding in clinical ophthalmology.
• The combination of anatomical and visual features offers a significant advancement in automated retinal image analysis.